Crop Type model has a new version

The SoilMate team has completed work on improving the results for the Crop Type model.
We used the method of combining two models to improve the quality of the CropType result.
In addition, we have made several changes to the model’s logic, increasing the input data by 1 month, and added small point filtering.

1. Change
Below are the results of the model from 06-01 to 08-31 and 05-01 08-31

It can be seen that the addition of one month improved the results, so we decided to also check if we add another 1 month.

We got even better results, but the time to get the results doubled compared to the period from 05-01 to 08-31. Therefore, we decided to stop at the golden mean.

2. Change
We added small point filtering when there is a clear error in the detection, now the model corrects it by itself, so the result becomes smoother

3. Change
And the main change is that we have integrated our Plot Boundary detection model into the logic of the Crop Type model so that the Crop Type model can see field boundaries more clearly and understand the monoculture of the field. Thanks to this, the model now works better with field and crop boundaries in the image, which provides greater accuracy in measuring the area and accuracy in determining the crop.
Below is a comparison of the Before and After results

Please check it out for yourself on our portal.soilmate.ai